What we shipped.

When Granite doesn't recognize a document type — eyeglass prescriptions, marriage certificates, HOA letters — it lands in Uncategorized. Open one of those documents and you'll see a "Help Granite learn this" card in the right rail. Send the file to our team and we'll add support for documents like it. Take it back any time from the same card.

Right-click a row in the Library, a collection or entity card, or a search result and you get the same actions the kebab menu offers — open, download, reprocess, rename, delete. Select a few rows first and the menu switches to bulk.

Drop a vet record, vaccination card, or microchip registration in your vault and Granite reads it as a pet record — patient name, species, breed, owner, provider, and date — instead of forcing it into a human-medical type. The pet itself becomes its own entity (with a paw icon, not a person icon), and an auto-collection like “Pet: Baxter” gathers everything about that pet in one place.

Cleaning up duplicates on /entities is now a drag. Grab a card and drop it onto another of the same kind — vendor onto vendor, person onto person — and Granite folds the first into the second. Same merge as the kebab menu, just faster when you can see both at once.

Image uploads now work the same as PDFs. Drag in a JPEG, PNG, WebP, or an iPhone HEIC photo and Granite reads it, classifies it, pulls out the fields, and files it.

If Granite created two entries for the same vendor or person — like “Comcast” and “Comcast Cable Communications” — you can now fold one into the other. Open the duplicate, hit the kebab menu, choose “Merge into…,” pick the survivor. All linked documents and collections move over.

Open any person on /entities and set a household role. Tagged people get pinned to a Household section at the top of Entities, and new documents that name them — even with extra middle names — snap to your tagged entity instead of creating a new one.

New documents now collapse name variants automatically. A first + last name match with or without middle names lands as a single person — the longest variant becomes the display name. Jr. and Sr. stay separate.

When a will, affidavit, or notarized form mentions a witness or notary public, Granite used to add that person to /entities. Now those incidental signers are skipped — only the document’s actual parties make it into your graph.

The document viewer now renders the real PDF instead of a flat image. Drag-select a paragraph, hit ⌘C, and the actual text comes with you. Scanned PDFs still display as images; selection works for any PDF with an embedded text layer.

A contract with a signing bonus and a monthly payment used to collapse to a single “total value” and pick the wrong one. Granite now extracts the one-time amount, the recurring amount, and the cadence as separate fields.

Granite already knew which contracts, policies, and IDs were about to expire — now you get a Monday-morning email of anything in the next 30 days. The digest carries titles and dates only, never extracted field values. Turn it off any time in Settings → Profile.

Drop the same PDF in twice and Granite says “already in vault” instead of silently re-processing it. The smarter half catches re-scans of the same logical document — your driver’s license photographed last year and again this month, the same W-2 saved from two different emails — and surfaces a quiet banner on the newer one with Keep both / Delete this one.

Pick one trusted person who can request a full copy of your archive if you’re ever unreachable. When they ask, you get a single email — Approve, Deny, or Wait — and you control how long the wait runs. An optional inactivity switch fires the grant automatically if you don’t sign in for a stretch of time.

Document titles and entity names from all-caps OCR sources — state forms, W-2 boxes, thermal receipts, ID cards — now render in normal title case (“Wal-Mart” instead of “WAL-MART”). Acronyms like LLC, IRS, and EIN stay capitalized.

Drop in prescriptions, lab results, visit summaries, EOBs, immunization records, and imaging reports — they classify into their own categories and get the right fields extracted. Refill counts and next-due dates surface as reminders.

Documents with low-quality embedded text (phone scanners, junk-character PDFs) now route through real OCR instead of trusting the broken text layer. When nothing is a confident match, docs land in “uncategorized” rather than being forced into a wrong fit.

Searching for 6789 now finds docs containing that digit run anywhere — last four of an SSN, the tail of an account number, a partial invoice ID. Full-text and tag-style queries are unchanged.

Docs that hang during ingest (worker restarts, OOMs, deploys) now self-heal within minutes. If a doc still looks stuck after 2 minutes, a Retry banner appears on the document page so you can kick it manually.

When a filename carries useful context that the extraction step missed — year/make/model, party names, document kind — those specifics now carry through to the final title instead of getting flattened to a generic label.

If you haven’t told us your name, the dashboard and toasts no longer pretend to know it. Add or update your name in Settings → Profile to personalize the greetings.